Class | PhusionPassenger::AbstractRequestHandler |
In: |
lib/phusion_passenger/abstract_request_handler.rb
|
Parent: | Object |
The request handler is the layer which connects Apache with the underlying application‘s request dispatcher (i.e. either Rails‘s Dispatcher class or Rack). The request handler‘s job is to process incoming HTTP requests using the currently loaded Ruby on Rails application. HTTP requests are forwarded to the request handler by the web server. HTTP responses generated by the RoR application are forwarded to the web server, which, in turn, sends the response back to the HTTP client.
AbstractRequestHandler is an abstract base class for easing the implementation of request handlers for Rails and Rack.
Some design decisions are made because we want to decrease system administrator maintenance overhead. These decisions are documented in this section.
Because only the web server communicates directly with a request handler, we want the request handler to exit if the web server has also exited. This is implemented by using a so-called _owner pipe_. The writable part of the pipe will be passed to the web server* via a Unix socket, and the web server will own that part of the pipe, while AbstractRequestHandler owns the readable part of the pipe. AbstractRequestHandler will continuously check whether the other side of the pipe has been closed. If so, then it knows that the web server has exited, and so the request handler will exit as well. This works even if the web server gets killed by SIGKILL.
Incoming "HTTP requests" are not true HTTP requests, i.e. their binary representation do not conform to RFC 2616. Instead, the request format is based on CGI, and is similar to that of SCGI.
The format consists of 3 parts:
HTTP headers are transformed to a format that satisfies the following grammar:
headers ::= header* header ::= name NUL value NUL name ::= notnull+ value ::= notnull+ notnull ::= "\x01" | "\x02" | "\x02" | ... | "\xFF" NUL = "\x00"
The web server transforms the HTTP request to the aforementioned format, and sends it to the request handler.
HARD_TERMINATION_SIGNAL | = | "SIGTERM" | Signal which will cause the Rails application to exit immediately. | |
SOFT_TERMINATION_SIGNAL | = | "SIGUSR1" | Signal which will cause the Rails application to exit as soon as it‘s done processing a request. | |
BACKLOG_SIZE | = | 500 | ||
MAX_HEADER_SIZE | = | 128 * 1024 | ||
OBJECT_SPACE_SUPPORTS_LIVE_OBJECTS | = | ObjectSpace.respond_to?(:live_objects) | ||
OBJECT_SPACE_SUPPORTS_ALLOCATED_OBJECTS | = | ObjectSpace.respond_to?(:allocated_objects) | ||
OBJECT_SPACE_SUPPORTS_COUNT_OBJECTS | = | ObjectSpace.respond_to?(:count_objects) | ||
GC_SUPPORTS_TIME | = | GC.respond_to?(:time) | ||
GC_SUPPORTS_CLEAR_STATS | = | GC.respond_to?(:clear_stats) |
connect_password | [RW] | A password with which clients must authenticate. Default is unauthenticated. |
iterations | [R] | The number of times the main loop has iterated so far. Mostly useful for unit test assertions. |
memory_limit | [RW] |
Specifies the maximum allowed memory usage, in MB. If after having
processed a request AbstractRequestHandler detects that
memory usage has risen above this limit, then it will gracefully exit (that
is, exit after having processed all pending requests).
A value of 0 (the default) indicates that there‘s no limit. |
processed_requests | [R] | Number of requests processed so far. This includes requests that raised exceptions. |
server_sockets | [R] |
A hash containing all server sockets that this request handler listens on.
The hash is in the form of:
{ name1 => [socket_address1, socket_type1, socket1], name2 => [socket_address2, socket_type2, socket2], ... } name is a Symbol. socket_addressx is the address of the socket, socket_typex is the socket‘s type (either ‘unix’ or ‘tcp’) and socketx is the actual socket IO objec. There‘s guaranteed to be at least one server socket, namely one with the name +:main+. |
soft_termination_linger_time | [RW] | If a soft termination signal was received, then the main loop will quit the given amount of seconds after the last time a connection was accepted. Defaults to 3 seconds. |
Clean up temporary stuff created by the request handler.
If the main loop was started by main_loop, then this method may only be called after the main loop has exited.
If the main loop was started by start_main_loop_thread, then this method may be called at any time, and it will stop the main loop thread.
Remove this request handler from the application pool so that no new connections will come in. Then make the main loop quit a few seconds after the last time a connection came in. This all is to ensure that no connections come in while we‘re shutting down.
May only be called while the main loop is running. May be called from any thread.